Add distributed training support to the Trainer

[arxyzan]

Add support for training a model on multiple GPUs. Methods and techniques will be discussed later in this issue.